As a special kind of linear system, the three-order block saddle point problem has challenging to study its iterative solution. Based on the classical generalized successive over relaxation (GSOR) method, the centered preconditioned GSOR method with three parameters for a class of three-order block large sparse saddle point problem is established and the convergence condition is discussed in this paper. Moreover, experimental results show that the new method has an advantage of computational cost over the centered preconditioned Uzawa-Low method. In addition, an extended one of the new method is provided, implementation details and analyses of corresponding framework about $i$-order block systems are shown, the blocking for saddle point problems are preliminarily proposed by some numerical results.